Claude and the First 3D Bacterial Vault: What Anthropic Mapped

A cathedral-sized shell with no known job

Vault particles are a longstanding mystery.

A vault particle is a large, hollow barrel of protein. Leonard Rome and Nancy Kedersha found it in 1986. Under an electron microscope the ribbed shell reminded them of a cathedral’s vaulted ceiling, which is where the name comes from. In mammals the shell is built from 78 copies of the major vault protein, arranged as two identical halves joined at the middle. At roughly three times the mass of a ribosome, it is one of the largest known molecular complexes in eukaryotes, the organisms with complex cells, including animals and plants.

Vaults sit in nearly all the cells of the human body and across most of the eukaryotic tree, yet nobody knows what they do. Anthropic sets that blank next to a wider pattern in cell biology: cells were first seen in 1665, and even in E. coli, one of the best-studied bacteria, more than a third of roughly 4,400 genes remain poorly understood.

The same shape, now in bacteria

We have shown that similar proteins from bacteria (and the viruses that infect them) can also assemble into vaults.

Until this work, vaults were known only in complex cells. Anthropic’s team reports that related proteins from bacteria, and from the viruses that infect those bacteria, can assemble into vaults as well. The write-up on anthropic.com calls the result the first 3D structure of a vault particle found in prokaryotic cells, the simpler, usually single-celled side of life. The hope is practical: a structure from a simpler organism may make the still-unknown role in human cells easier to test.

Claude was used at each research step the page describes: searching sequence literature, optimizing a method for isolating the particles, and mapping the structure. Anthropic says a preprint has been submitted to bioRxiv and that the link on the page would be updated once it is live. The public account of the project is the article on anthropic.com.

Days of work, and the errors in between

Manually determining the vault particle’s structure would have taken weeks to months of painstaking work, but here it took days.

Anthropic’s life sciences team treats the project as a demonstration of both the potential and the limits of using AI to speed science. Determining the structure by hand would have taken weeks to months. Working with Claude compressed that stage to days. Along the way Claude made simple mistakes. The biology team worked with Anthropic’s machine learning researchers to address them.

The open scientific question is unchanged: what a vault does inside a cell is still unknown. The new map is a closer structural clue, from a form of life that is easier to study, rather than an answer to the function itself.

• Literature search

  Claude helped search across sequence literature while the team looked for vault-related proteins.

• Isolation

  It helped optimize a method for isolating the particles before the structure could be mapped.

• Structure mapping

  The 3D map itself is the stage Anthropic says shrank from weeks or months to days.

• Human correction

  Simple model mistakes were caught and fixed with the machine learning researchers, so the speed-up was a collaboration.